Marcus Vale: Okay, this one got me. Genuinely. June 11 — Anthropic drops Claude Corps, a hundred and fifty million dollars, a thousand fellows, eighty-five grand salary plus benefits, embeds them inside four hundred nonprofits. RAINN, Goodwill Industries International, Code for America. And they're calling it — I'm reading this right — a response to AI-driven labor displacement.
Ben Okonkwo: The timing is — yeah, go on.
Marcus Vale: The timing is the whole thing. Tech is averaging nine hundred and thirty-five layoffs per day this year. Nine-thirty-five. Per day. Anthropic's answer is one thousand twelve-month fellowships. That's — I mean, that's not a solution, that's a press release wearing a solution's clothing. This is the most expensive brand-awareness campaign in nonprofit history, and they filed it under social responsibility.
Ben Okonkwo: Now, I'd push back on brand-awareness — actually, wait. The commercial logic here is pretty direct. You embed a thousand Claude-trained workers inside four hundred institutions, those institutions build workflows around Claude. That's a deployment pipeline, not a donation.
Marcus Vale: Right. And then the punchline — Dario Amodei publishes his essay the same week. June 10, 11. Calling for mandatory government authority to block frontier AI models. And within days, the government blocked his models. BIS, export controls, the whole thing. He basically wrote the template for his own restriction.
Ben Okonkwo: That's the part I can't stop pulling at. Did he invite that?
Ben Okonkwo: No — actually, that's the thing I want to slow down on. Because the mechanism doesn't match. At all. Amodei's essay calls for mandatory audits triggered by quantitative thresholds — ten-to-the-twenty-five FLOPs of training compute, or five hundred million dollars in annual AI revenue. Independent third-party testing. Published findings. That's what he proposed. What actually happened was — the Bureau of Industry and Security issued a mandatory export control on June 12th, based on security concerns attributed to Amazon's CEO. No published technical justification. No capability threshold. No independent validation.
Marcus Vale: Hold on. Amazon's CEO.
Ben Okonkwo: Amazon — Anthropic's largest investor via AWS — their CEO raises the concern that triggers BIS to move. And then Anthropic gets summoned to the White House. That's — I mean, that sequence is genuinely strange. It's not what Amodei asked for. He wanted audits. He got a verbal security report and a trade restriction.
Marcus Vale: That's — yeah. That's a different thing entirely.
Ben Okonkwo: And then layer NSPM-11 on top. The memo accelerates AI adoption in national security contexts but broadly excludes Anthropic from preferred vendor status — except it carves out Claude Mythos, which is already deployed at the NSA. So Anthropic is simultaneously restricted and essential. The government's position is almost contradictory.
Marcus Vale: Wait — so Trump's June 2nd executive order is explicitly voluntary, can't be read as authorizing mandatory regulation, and then ten days later BIS issues a mandatory restriction. That's not the framework Amodei was lobbying for. That's the government just — doing what it wants.
Marcus Vale: But wait — the waiver. That's the tell. NSPM-11 carves out Claude Mythos specifically. The NSA is running Claude Mythos right now. Under that waiver. And simultaneously, BIS is restricting international access to Anthropic's models. That's — the government is not acting adversarially here, that's co-dependence.
Ben Okonkwo: Yeah. That structure is genuinely strange.
Marcus Vale: It's one buyer. The national security apparatus. Anthropic has leverage over exactly one buyer and zero commercial optionality internationally — because the export control just killed that. So the waiver isn't a gift, it's — I mean, it's almost a trap? You're essential to the NSA and captive to it at the same time.
Ben Okonkwo: Now that's — actually, okay, I'll grant that. The bifurcated posture is real. Restricted everywhere, indispensable inside Fort Meade.
Marcus Vale: And then Amodei's thresholds. Ten-to-the-twenty-five FLOPs or five hundred million in annual AI revenue — who does that exclude? Not Anthropic. Anthropic's compliance infrastructure becomes the differentiator. Smaller competitors can't absorb that audit cost. That's a moat dressed as a safety proposal.
Ben Okonkwo: Which is — and I want to flag this, because it's almost too convenient to be dismissed — Anthropic's own Claude Corps announcement literally says the companies building this technology have a responsibility to invest directly in the workers absorbing the change. That framing is simultaneously sincere and a very clean commercial argument for why Anthropic gets to write the rules.
Ben Okonkwo: And that's — I mean, that's the thing I want to land on before we close. Because the precedent Amodei actually created isn't what he proposed. He called for independent audits, published thresholds, the Export Administration Regulations framework applied systematically — the way BIS handles semiconductors, encryption tech, stuff with decades of legal infrastructure. What actually happened is a verbal security report from Amazon's CEO was sufficient to trigger a global restriction on a frontier AI lab. That template — that's not Anthropic's. Any administration can use it. On anyone.
Marcus Vale: Be careful what you wish for, I guess.
Ben Okonkwo: Yeah, but — it's more specific than that. The government didn't audit Anthropic and find it dangerous. They didn't hit the ten-to-the-twenty-five FLOPs threshold, didn't run the independent validation Amodei actually asked for. One investor-CEO made a phone call. That's what moved BIS.
Marcus Vale: And Dario's at the White House explaining himself to the people who own the only waiver keeping Claude Mythos inside the NSA.
